AI Assistant

data-peek includes a powerful AI assistant that helps you write SQL queries, generate charts, and get insights from your data.

Getting Started

Open the AI Panel

• Press Cmd+I (macOS) or Ctrl+I (Windows/Linux)
• Or click the sparkle icon in the toolbar
• Or use the command palette (Cmd+K) and search for “AI”

Choose a provider

Open AI Settings (from the AI panel or the command palette) and pick one provider. The app always uses whichever provider is marked Active — you can switch anytime.

• Claude Code (local CLI) — no API key. If you have the claude CLI installed and signed in, select the tile and click Use Claude Code. That single click makes it active — nothing else to configure. See Bring your own harness below.
• API-key providers (OpenAI, Anthropic, Google, …) — paste your key, click Save … Key, then Set Active. Keys are stored in your OS keychain and only ever sent to that provider.
• Ollama — local models, no key; point it at your Ollama URL and click Use Ollama.

To use only the local Claude CLI, just make Claude Code (local CLI) the active provider — the others can stay configured but unused.

Supported Providers

Note: Available models may change as providers release updates. Check your provider’s documentation for the latest model offerings.

Bring your own harness (Claude Code)

New in v0.27: instead of pasting an API key, point the assistant at the claude CLI you already have installed. data-peek drives your local Claude Code, on your own login — so there’s nothing to store. And when the MCP server is enabled, the assistant runs grounded: it queries your real database (read-only, always rolled back) to verify its answer before replying.

1. Install the Claude Code CLI and sign in (run claude once).
2. In AI Settings, choose Claude Code (local CLI) — no API key required.
3. Enable the MCP server (Settings → MCP) so the assistant can query your database.
4. Ask in plain English. The reply streams in live — you see the answer as it’s written, and while it grounds itself you get a running indicator of each step (“Running query…”, “Reading schema…”). Read-only queries run automatically, and grounded answers show a “Grounded against your database” badge.

New in v0.28

The assistant now streams every reply, remembers the conversation across turns, and can build more from a single prompt.

Streaming, grounded answers with memory

Answers stream token-by-token with a live grounding indicator, and follow-up questions (“now just the top 3 of those”) continue the same conversation.

Mini-reports and @ mentions

Ask for an overview and get an inline mini-dashboard — KPIs, charts, and a table — from a single prompt, each widget backed by a verified read-only query. Type @ in the composer to autocomplete real tables and columns.

“Give me a business overview: key totals plus a couple of charts”

More

• Follow-up chips — the assistant suggests your next question; click one to ask it.
• Pin from chat — save a generated query, or add it to a dashboard or notebook cell, straight from the answer.

Features

Natural Language to SQL

Ask questions in plain English and get SQL queries:

“Show me all users who signed up in the last 30 days”

The AI generates:

SELECT * FROM users
WHERE created_at >= NOW() - INTERVAL '30 days'
ORDER BY created_at DESC;

Schema-Aware Queries

The AI assistant understands your database schema. It knows your:

• Table names and structures
• Column names and types
• Relationships between tables

This means you can ask questions naturally:

“How many orders does each customer have?”

Chart Generation

Generate visualizations directly from your queries:

“Show me a bar chart of monthly revenue for 2024”

The AI will:

1. Generate the appropriate SQL query
2. Execute it against your database
3. Create an interactive chart

Supported chart types:

• Bar charts
• Line charts
• Pie charts
• Area charts

Query Explanations

Get explanations for complex queries:

“Explain what this query does”

Paste any SQL and the AI will break down:

• What each part does
• How the joins work
• What the result will look like

Safety Guardrails

The AI assistant includes safety features:

• Destructive query warnings: DELETE, DROP, TRUNCATE queries show a confirmation dialog
• Read-only by default: Queries are analyzed before execution
• Schema access only: The AI can see schema structure but not your actual data

Usage Tips

Be Specific

More specific questions get better results:

Reference Tables by Name

When your question could apply to multiple tables, be explicit:

“Show me the status distribution of the orders table”

Iterate on Results

You can refine queries in conversation:

1. “Show me all products”
2. “Filter to only in-stock items”
3. “Add the category name from the categories table”

Keyboard Shortcuts

Privacy & Security

• BYOK model: You use your own API key, so queries go directly to your chosen provider
• Schema context: Only schema metadata (table/column names and types) is sent to the AI, not your actual data
• Local storage: Conversation history is stored locally on your device
• No telemetry: data-peek doesn’t collect or transmit your queries or results
• Provider errors: Error messages returned by AI providers are generated by those providers and may include contextual information about your request